
import cv2
import numpy as np
from matplotlib import pyplot as plt
from csv import reader
import pandas as pd
from tensorflow import keras
from tensorflow.keras import applications
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras import optimizers
from tensorflow.keras.models import Sequential, Model 
from tensorflow.keras.layers import *
from tensorflow.keras.callbacks import ModelCheckpoint, LearningRateScheduler, TensorBoard, EarlyStopping
import os
from sklearn.model_selection import train_test_split
import tensorflow.keras.metrics as km
from sklearn.metrics import accuracy_score
from sklearn.metrics import precision_score
from sklearn.metrics import recall_score
from sklearn.metrics import f1_score
from sklearn.metrics import cohen_kappa_score
from sklearn.metrics import roc_auc_score
from sklearn.metrics import multilabel_confusion_matrix
from sklearn.metrics import classification_report


#build dataset for ML model training

X = []
Y = []

color_line = (255, 255, 255)
thickness = 10
img_height , img_width = 400, 100

previous_group = "empty"
image=np.zeros((1000,4000,3),np.uint8)
previous_point = (0,0)
frames_list = []

with open('ftir_metadata.csv', 'r') as read_obj_classes:
    csv_reader_classes = reader(read_obj_classes)
    header = next(csv_reader_classes)
    classes = []
    for row in csv_reader_classes:
      if row[0] not in classes:
        classes.append(row[0])

with open('ftir_library.csv', 'r') as read_obj:
    csv_reader = reader(read_obj)
    header = next(csv_reader)
    sortedlist = sorted(csv_reader, key=lambda row: int(row[3]))
    df=pd.read_csv("ftir_metadata.csv")

    # Iterate over each row in the csv using reader object
    for row in sortedlist:
        group = row[3]
        point = (int(float(row[0])), int(float(row[1])*1000))

        if previous_group == "empty":
            image=np.zeros((1000,4000,3),np.uint8)
            frames_list = []
            previous_group = group            
        elif (group != previous_group):
            y = [0]*len(classes)     
            val = df.iloc[int(previous_group)-1]['spectrum_identity']
            index = classes.index(val)
            y[index] = 1
            Y.append(y)

            print(previous_group)
            print(classes[index])
            plt.figure(figsize=(20, 20))
            plt.imshow(image, origin='lower')
            plt.show()

            image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
            image = cv2.resize(image, (img_height, img_width))
            frames_list.append(image)
            X.append(frames_list)

            frames_list = []
            image=np.zeros((1000,4000,3),np.uint8)

            previous_group = group
        else:           
            image = cv2.line(image, previous_point, point, color_line, thickness)  

        previous_point = point

    y = [0]*len(classes)     
    val = df.iloc[int(previous_group)-1]['spectrum_identity']
    index = classes.index(val)
    y[index] = 1
    Y.append(y)  

    print(previous_group)
    print(classes[index])
    plt.figure(figsize=(20, 20))
    plt.imshow(image, origin='lower')
    plt.show()

    image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    image = cv2.resize(image, (img_height, img_width))        

    frames_list.append(image)
    X.append(frames_list)

1
HDPE

2
HDPE

3
HDPE

4
HDPE

5
HDPE

6
HDPE

7
HDPE

8
HDPE

9
HDPE

10
HDPE


X = np.asarray(X)
Y = np.asarray(Y)

X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=0.30, shuffle=True, random_state=0)
 
model = Sequential()
model.add(ConvLSTM2D(filters = 64, kernel_size = (3, 3), return_sequences = False, data_format = "channels_last", input_shape = (1, 100, 400, 1)))
model.add(Dropout(0.2))
model.add(Flatten())
model.add(Dense(128, activation="relu"))
model.add(Dropout(0.3))
model.add(Dense(len(classes), activation = "softmax"))
 
model.summary()
 
opt = keras.optimizers.SGD(lr=0.001)
model.compile(loss='categorical_crossentropy', optimizer=opt, metrics=["accuracy"])
 
earlystop = EarlyStopping(patience=5)
callbacks = [earlystop]
 
history = model.fit(x = X_train, y = y_train, epochs=15, batch_size = 8 , shuffle=True, validation_split=0.20, callbacks=callbacks)

Model: "sequential_1"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv_lstm2d_1 (ConvLSTM2D)  (None, 98, 398, 64)       150016    
                                                                 
 dropout_2 (Dropout)         (None, 98, 398, 64)       0         
                                                                 
 flatten_1 (Flatten)         (None, 2496256)           0         
                                                                 
 dense_2 (Dense)             (None, 128)               319520896 
                                                                 
 dropout_3 (Dropout)         (None, 128)               0         
                                                                 
 dense_3 (Dense)             (None, 251)               32379     
                                                                 
=================================================================
Total params: 319,703,291
Trainable params: 319,703,291
Non-trainable params: 0
_________________________________________________________________
Epoch 1/15

/usr/local/lib/python3.7/dist-packages/keras/optimizer_v2/gradient_descent.py:102: UserWarning: The `lr` argument is deprecated, use `learning_rate` instead.
  super(SGD, self).__init__(name, **kwargs)

45/45 [==============================] - 10s 175ms/step - loss: 3.7610 - accuracy: 0.4073 - val_loss: 2.9975 - val_accuracy: 0.4719
Epoch 2/15
45/45 [==============================] - 8s 168ms/step - loss: 2.8015 - accuracy: 0.5140 - val_loss: 2.8397 - val_accuracy: 0.5169
Epoch 3/15
45/45 [==============================] - 8s 169ms/step - loss: 2.2699 - accuracy: 0.6545 - val_loss: 2.7217 - val_accuracy: 0.5169
Epoch 4/15
45/45 [==============================] - 8s 169ms/step - loss: 1.7160 - accuracy: 0.7584 - val_loss: 2.6833 - val_accuracy: 0.5506
Epoch 5/15
45/45 [==============================] - 8s 168ms/step - loss: 1.2854 - accuracy: 0.8539 - val_loss: 2.6580 - val_accuracy: 0.5730
Epoch 6/15
45/45 [==============================] - 8s 168ms/step - loss: 0.9787 - accuracy: 0.8904 - val_loss: 2.6630 - val_accuracy: 0.5730
Epoch 7/15
45/45 [==============================] - 8s 168ms/step - loss: 0.7155 - accuracy: 0.9326 - val_loss: 2.6417 - val_accuracy: 0.5618
Epoch 8/15
45/45 [==============================] - 8s 177ms/step - loss: 0.5775 - accuracy: 0.9354 - val_loss: 2.6375 - val_accuracy: 0.5506
Epoch 9/15
45/45 [==============================] - 8s 169ms/step - loss: 0.4808 - accuracy: 0.9438 - val_loss: 2.6341 - val_accuracy: 0.5618
Epoch 10/15
45/45 [==============================] - 8s 169ms/step - loss: 0.3638 - accuracy: 0.9663 - val_loss: 2.6536 - val_accuracy: 0.5730
Epoch 11/15
45/45 [==============================] - 8s 168ms/step - loss: 0.3000 - accuracy: 0.9635 - val_loss: 2.6596 - val_accuracy: 0.5730
Epoch 12/15
45/45 [==============================] - 8s 168ms/step - loss: 0.2456 - accuracy: 0.9747 - val_loss: 2.6466 - val_accuracy: 0.5843
Epoch 13/15
45/45 [==============================] - 8s 168ms/step - loss: 0.2389 - accuracy: 0.9691 - val_loss: 2.6531 - val_accuracy: 0.5843
Epoch 14/15
45/45 [==============================] - 8s 168ms/step - loss: 0.1961 - accuracy: 0.9803 - val_loss: 2.6548 - val_accuracy: 0.5843


# Calling `save('my_model')` creates a SavedModel folder `my_model`.
model.save("my_model")

INFO:tensorflow:Assets written to: my_model/assets